Search papers, labs, and topics across Lattice.
The paper introduces a deep clustering-based boundary-decoder net (DCBD) for predicting inter- and intra-layer stress in heterogeneous integrated IC chips under thermal cycling. The method addresses the ill-posed nature of boundary-decoder nets by incorporating deep clustering to map material parameters to a shared latent space with corresponding stress images. Experiments on a simulated IC chip dataset demonstrate that the proposed DCBD outperforms baseline and BD-net variants in both training and testing error reduction for stress prediction.
By combining deep clustering with a boundary-decoder network, this work achieves more accurate stress prediction in heterogeneous IC chips compared to existing methods.
High stress occurs when 3D heterogeneous IC packages are subjected to thermal cycling at extreme temperatures. Stress mainly occurs at the interface between different materials. We investigate stress image using latent space representation which is based on using deep generative model (DGM). However, most DGM approaches are unsupervised, meaning they resort to image pairing (input and output) to train DGM. Instead, we rely on a recent boundary-decoder (BD) net, which uses boundary condition and image pairing for stress modeling. The boundary net maps material parameters to the latent space co-shared by its image counterpart. Because such a setup is dimensionally wise ill-posed, we further couple BD net with deep clustering. To access the performance of our proposed method, we simulate an IC chip dataset comprising of 1825 stress images. We compare our new approach using variants of BD net as well as a baseline approach. We show that our approach is able to outperform all the comparison in terms of train and test error reduction.
